Dynamic interaction of Syntactic and Semantic Analyses by Equivalent Transformation

摘要

In conventional natural language processing, syntactic and semantic analyses are executed sequentially. The sequential method is, however, inefficient due to the inflexibility of the order of computation. In this paper, we present a system design method based on equivalent transformation to construct a natural language understanding system, which executes integrated processing of syntactic and semantic analyses. The use of equivalent transformation has the additional benefit that the analysis is provably correct and complete. Equivalent transformation changes a declarative description of the problem one step at a time and each step is proven to preserve the semantic meaning of the program. In this system, many constraints are used for description of problems and the processing order is decided, not at compile time, but dynamically at execution time depending on the current constraints. The benefits of using the equivalent transformation system include speed, provable correctness, reliability, and intuitive design.